AI-Powered Microscopy Reveals Hidden Drug Effects Through Nucleolus Shape Changes
核心洞察
Princeton researchers developed an AI tool that classifies nucleolus shapes to reveal how drugs affect cellular structures at the single-cell level.
The neural network identified three known nucleolar morphologies and discovered an entirely new "flower" shape induced by the chemotherapy drug topotecan.
Two anti-cancer (搜索) drugs were found to cause "cap" formations, a phenomenon not previously reported, suggesting unrecognized effects on nucleolar function.
A team at Princeton University (搜索) has developed an artificial intelligence-driven imaging platform capable of decoding how drugs reshape the nucleolus—a critical biomolecular condensate inside cells—and in doing so has uncovered a previously unknown structural signature triggered by a widely studied chemotherapy agent. The work, published June 4 in Cell, introduces a tool that could transform how researchers screen and evaluate drug effects at the single-cell level.
The study, led by Cliff Brangwynne, the June K. Wu '92 Professor of Chemical and Biological Engineering and director of the Omenn-Darling Bioengineering Institute, focused on biomolecular condensates: tiny membraneless droplets that orchestrate transcription and gene regulation and have been implicated in diseases including Alzheimer's (搜索), ALS (搜索), and cancer (搜索).
"The central problem in biology is how do you get emergent structure from individual molecular interactions," said Brangwynne. "The key innovation here was to develop a way to learn from the images and classify the patterns that are emergent."
Training a Neural Network to See the Invisible
Postdoctoral researcher Anita Donlic and research software engineer Troy Comi trained a neural network on tens of thousands of images of cells containing healthy, spherical nucleoli alongside two well-known atypical forms: a "cap" shape and a "beaded necklace" shape. These morphologies have been linked to distinct cellular stress responses. Cap shapes can arise from treatments that disrupt RNA production needed to assemble protein-making machinery, while necklace shapes can result from drugs that interfere with a separate RNA-related process.
The team used advanced microscopy to image nucleolar shape changes in hundreds of human cells under a range of drug-controlled conditions. The resulting images—difficult to interpret even for highly trained scientists—were fed through the machine-learning tool, which sorted them into four basic categories: three that the researchers expected and a fourth that was entirely unexpected.
A New Morphology Emerges
For the chemotherapy drug topotecan, the neural network discovered a nucleolus shape that did not fit any known category. The researchers labeled it "flower."
"No one's seen this flower morphology before," Brangwynne said. "The network flagged it as not fitting neatly into the other three categories."
Topotecan was already known to inhibit TOP1 (搜索), an enzyme involved in DNA replication. Donlic demonstrated that loss of TOP1 induced the flower shape, revealing the enzyme's previously unappreciated role in maintaining nucleolar organization through regulation of RNA processing.
Unmasking Hidden Drug Effects
The neural network also found that two known anti-cancer (搜索) drugs caused cap formations—a phenomenon not previously reported for those agents. This finding suggests these drugs may be affecting nucleolar function in ways that had gone unrecognized, according to Donlic, the paper's first author.
Across the drug panel, the team observed that different concentrations of drugs produced different degrees of change in both cap and necklace morphologies, pointing toward a dose-dependent relationship between drug exposure and condensate remodeling.
Beyond the Nucleolus
The researchers extended their approach to other condensates linked to RNA processes. They observed similar dose-and-response results for drugs targeting nuclear speckles—hubs for messenger RNA activity—and for condensates formed by respiratory syncytial virus. This broader applicability underscores the platform's potential as a generalizable system for monitoring cellular responses to pharmacologic perturbations.
"You could be missing other important features," Donlic concluded. "Things that could tell you there's new biology."
The findings establish a foundation for a robust, image-based system to evaluate drug effects at single-cell resolution, potentially accelerating the identification of therapeutic candidates and revealing off-target activities that conventional assays might overlook.
The paper, "Deep Learning of Functional Perturbations from Condensate Morphology," includes co-authors Sofia Quinodoz, Nima Jaberi-Lashkari, Krist Antunes Fernandez, Lifei Jiang, Lennard Wiesner, and Ai Ing Lim of Princeton University (搜索). Support was provided in part by the Howard Hughes Medical Institute, the Princeton Center for Complex Materials, the St. Jude Collaborative on Membraneless Organelles, the Chan Zuckerberg Initiative Exploratory Cell Network, and the Princeton Laboratory for Artificial Intelligence.